Learning Dynamic Relationships for 3D Human Motion Prediction

Cui, Qiongjie; Sun, Huaijiang; Yang, Fei

doi:10.1109/cvpr42600.2020.00655

Cited by 132 publications

(120 citation statements)

References 13 publications

Supporting

Mentioning

120

Contrasting

Order By: Relevance

“…Most existing works [8,16,25,28] represent the pose as joints along a kinematic graph with joint orientations parameterized as axis angles. A related approach characterizes the skeletal structure as a graph [7,12,17,26,27]. A critical issue is that these schemes usually treat the joints on an equal standing, failing to account for the fact that the kinematic chain is a hierarchical structure.…”

Section: Skeleton-based Pose Representationmentioning

confidence: 99%

“…Explicit Dependency Modeling Effectively capturing the precise spatial relationships between closely related joints is indispensable to understanding and modeling motion. Existing methods have expended significant efforts in this direction such as considering kinematic trees [25] or graphs [7,17,26] and examining the problem in the frequency domain via discrete Fourier transforms [27]. Yet, the fixed graph or kinematic tree structures in such methods make it difficult to incorporate prior knowledge, while using frequency domain usually involves some intricate or cumbersome network designs to extract joint relationships.…”

Section: The Proposed Networkmentioning

confidence: 99%

“…H3.6m We benchmark the proposed method against 8 existing works, including Res-GRU [28], ConSeq2Seq [16], HMR [25], FC-GCN [27], TrajNet [22], SDMTL [21], LDR [7], and HRI [26]. The experimental results are demonstrated in Table 1 and Fig.…”

Section: Evaluation On Human Datasetsmentioning

confidence: 99%

See 2 more Smart Citations

Motion Prediction via Joint Dependency Modeling in Phase Space

Liu

et al. 2021

Proceedings of the 29th ACM International Conference on Multimedia

View full text Add to dashboard Cite

Motion prediction is a classic problem in computer vision, which aims at forecasting future motion given the observed pose sequence. Various deep learning models have been proposed, achieving stateof-the-art performance on motion prediction. However, existing methods typically focus on modeling temporal dynamics in the pose space. Unfortunately, the complicated and high dimensionality nature of human motion brings inherent challenges for dynamic context capturing. Therefore, we move away from the conventional pose based representation and present a novel approach employing a phase space trajectory representation of individual joints. Moreover, current methods tend to only consider the dependencies between physically connected joints. In this paper, we introduce a novel convolutional neural model to effectively leverage explicit prior knowledge of motion anatomy, and simultaneously capture both spatial and temporal information of joint trajectory dynamics. We then propose a global optimization module that learns the implicit relationships between individual joint features.Empirically, our method is evaluated on large-scale 3D human motion benchmark datasets (i.e., Human3.6M, CMU MoCap). These results demonstrate that our method sets the new state-of-the-art on the benchmark datasets. Our code is released at https://github. com/Pose-Group/TEID. CCS CONCEPTS• Computing methodologies → Computer vision; Activity recognition and understanding.

show abstract

Section: Skeleton-based Pose Representationmentioning

confidence: 99%

Section: The Proposed Networkmentioning

confidence: 99%

See 1 more Smart Citation

Motion Prediction via Joint Dependency Modeling in Phase Space

Liu

et al. 2021

Proceedings of the 29th ACM International Conference on Multimedia

View full text Add to dashboard Cite

show abstract

“…The human motion sequence is essentially a combination of 3D skeleton poses, and human joints provide meaningful connections in Cartesian space. Therefore, instead of CNNs or RNNs, we propose to exploit GCNs to extract these topological relationships [11,22]. We consider that an untrained GCN model can effectively extract some abstract prior without any training or pre-training procedure and action dataset.…”

Section: Deep Human Dynamics Priormentioning

confidence: 99%

“…For 3D skeleton-based human actions, it naturally presents a meaningful topological structure [22,30]. Hence, in contrast to CNNs, we propose to utilize GCNs to model the pairwise relations of human joints, which is conducive to efficiently access the skeleton relationship and preserve the detailed information of human dynamics [11,31,37]. We show that an untrained GCN with randomly initialized weights is capable of replacing the manually-design prior and has achieved remarkable performance in several representative motion reconstruction tasks.…”

Section: Introductionmentioning

confidence: 99%